Nervos Network (CKB): a complete guide to the coin and its mining

Nervos Network is a proof of work blockchain whose base layer is called Common Knowledge Base, CKB for short. The mainnet, code named Lina, went live on November 16, 2019. It was built by Cryptape and the Nervos Foundation, with Jan Xie, Daniel Lv, Terry Tai, Kevin Wang and Cipher Wang among the founders. The Rylai testnet had been running since May 2019.

The design brief differed from a general purpose blockchain from day one. Layer one here is deliberately slow and expensive in terms of space: it is a store of value and state layer, while computation and throughput are pushed to layer two. Hence the defining trait of the economics: a CKByte is not only money, it is the right to occupy one byte of network state. Keep data on layer one and you must hold the matching amount of coins.

Since 2024 the network has carried a second positioning: a layer 2 for Bitcoin. The RGB++ protocol binds Bitcoin UTXOs to CKB cells isomorphically, with no bridge and no wrapped tokens, while the Fiber channel network adds instant payments. That revived development, but so far it has left no visible mark on price or hashrate: by the summer of 2026 both sit at multi year lows.

This page is written as a reference for hardware owners. Below: the Eaglesong algorithm and NC-MAX consensus, the mechanics of dual issuance and NervosDAO, a ranking of all eight Eaglesong machines in our catalogue with honest electricity thresholds, pools, wallets, the ecosystem and a separate list of figures that the big coin directories still publish in outdated form.

• 2019The Rylai testnet goes up in May. On July 18 RFC 0010 is published with the description of the Eaglesong hash function, authored by Alan Szepieniec. On November 16 the Lina mainnet launches. Mining starts on GPUs and CPUs, no dedicated hardware exists yet.
• 2020The first Eaglesong ASICs arrive. In March the Todek Toddminer C1 ships at 1.55 TH/s, in April Bitmain releases the Antminer K5 at 1.13 TH/s, in November iBeLink shows the BM-N1 at 6.6 TH/s. Within months GPUs stop making sense on this algorithm.
• 2021In March the Goldshell CK5 arrives at 12 TH/s, in the summer the compact CK-BOX follows, and in December the CK6 at 19.3 TH/s. Goldshell effectively takes over the small and mid size niche on Eaglesong and still holds it.
• 2021On March 31 CKB posts its all time high, roughly 0.0437 dollars. From that mark the coin has lost more than 97 percent by 2026.
• 2022On May 10 the first hard fork activates on mainnet, the Mirana edition, epoch 5414. It brings CKB-VM version 1 with support for the B extension of RISC-V, macro operation fusion and an extension field in the block header for future work. The testnet had received the upgrade earlier, on November 24, 2021.
• 2023In January the Antminer K7 ships at 63.5 TH/s and 3,080 W. This is a generation gap: the machine is roughly three and a half times more efficient than anything on the market before it, and network hashrate climbs sharply.
• 2023On November 19 the first halving takes place at epoch 8,760. The epoch reward drops from 1,917,808 to 958,904 CKB, annual base issuance from 4.2 to 2.1 billion CKB. Secondary issuance is untouched.
• 2024On February 13 Cipher Wang presents the RGB++ protocol: Bitcoin UTXOs are bound isomorphically to CKB cells, which gives Bitcoin assets full smart contracts without a bridge. Developer tooling and the first wallets for the protocol follow through the year.
• 2024The application layer grows: the Spore protocol and DOB digital objects, stored entirely in network state with no external files, plus the modular UTXO Stack platform for assembling layer two solutions on top of Bitcoin.
• 2024In April the iBeLink BM-N3 Max ships at 30 TH/s. It is the last new Eaglesong machine: from then to the end of July 2026 no further models have reached the market. In September the Fiber payment channel testnet starts.
• 2024On October 25 the second protocol edition, named Meepo, goes live on testnet at epoch 9,690. It adds the Spawn syscall for direct exchange between scripts, the data2 value in the hash_type field and CKB-VM cycle optimisations.
• 2025On July 1 the Meepo edition activates on mainnet at epoch 12,293. That date, not the year 2024, separates the old network behaviour from the new one, even though the edition is named after the year it was prepared in.
• 2025Work on the Fiber channel network continues all year, and by December the testnet reaches version 0.6.0 with a plan for a mainnet release. In parallel the network explorer and the developer tooling get updates.
• 2026By late July the network hashrate falls to roughly 70 to 85 PH/s against hundreds of PH/s in 2024, and the price sets a new all time low. The second halving is expected in November 2027 at epoch 17,520.

The network stores no account balances, it stores a set of cells. A Cell is a generalised Bitcoin UTXO: it carries a capacity field, arbitrary data, a lock script and an optional type script. A cell is immutable. To change state, a transaction spends old cells and creates new ones, and nodes only verify that the script rules were respected.

The key economic detail hides in the capacity field. One CKByte is the right to occupy one byte of network state, and a cell physically cannot weigh more than its capacity. Metadata counts too: the lock and type scripts and the capacity field itself take up room. That is why even an empty working cell requires several dozen CKBytes, and every asset on this network carries a measurable storage cost.

A transaction is confirmed in two steps. First a shortened transaction hash enters a block, this is the proposal, and only several blocks later does the transaction itself land in a block as a commitment. The point is that by the time of commitment the transaction contents have already spread across the network, so propagating a new block does not depend on shipping fresh data. That keeps the block interval down to single digit seconds without raising the orphan rate.

Computation is executed by CKB-VM, a virtual machine built on the real RISC-V instruction set. It is not a bespoke bytecode but an industry architecture, so scripts are written in Rust and C and compiled with ordinary toolchains. Script complexity is bounded by a cycle counter rather than by a hard wired list of operations, which makes it possible to add new cryptography without changing the protocol itself.

1. The transaction is assembled from cellsThe wallet picks live cells that cover both the amount being sent and the capacity of the new cells. The total input capacity must be at least the total output capacity, and the difference becomes the fee.
2. The transaction reaches the node poolNodes check basic validity and hold the transaction locally. At this stage it is not yet considered accepted by the network.
3. The miner publishes a proposalA shortened transaction hash lands in the proposal zone of the next block. The protocol proposal window spans blocks two through ten counted from the current one.
4. The transaction is committed in a blockSeveral blocks later the miner includes the transaction itself in the commitment zone. Only now does network state change: old cells become dead, new ones become live.
5. Nodes execute the scriptsEvery cell is checked by its lock script and, where present, by its type script. Execution runs in CKB-VM on RISC-V, and spending is bounded by a cycle counter.
6. The miner collects the rewardThe payout combines base issuance, a share of secondary issuance, 40 percent of the fees for proposed transactions and 60 percent for committed ones.
7. The epoch closes and difficulty is recalculatedAn epoch lasts a fixed 4 hours. At the end of it the protocol looks at the orphan rate and adjusts difficulty so as to hold that rate near 2.5 percent. The number of blocks in the next epoch may change.

Eaglesong is a standalone hash function written specifically for this network. RFC 0010 is signed by Alan Szepieniec and dated July 18, 2019. The construction is a sponge: rate 256 bits, capacity 256 bits, output 256 bits, 43 rounds. Inside there are only addition, rotation and exclusive or operations over 32 bit words, which is why such designs are called ARX. Each round consists of multiplication by a bit matrix, circular multiplication with rotation coefficients, addition of a round constant and an add, rotate, add block.

Picking a new function was a deliberate choice. The developers wanted neither to inherit SHA-256 with its ready made hardware fleet, nor to build an ASIC resistant algorithm that would lose the fight sooner or later. The logic ran like this: offer a function that is simple and clear to implement in silicon, so that dedicated hardware appears quickly and from several vendors at once instead of falling to a single player. In practice that is what happened: the first ASICs arrived in 2020, only months after the network launched.

An ARX construction is very cheap in hardware and needs almost no memory. That is exactly why GPUs lost their purpose on Eaglesong earlier than on most other algorithms: a graphics card holds no advantage at all where the whole job reduces to simple integer operations with a high degree of parallelism. CPU mining is even more pointless. Since 2020 this has been a pure ASIC algorithm.

On top of the hash function runs the NC-MAX consensus described in RFC 0020. It differs from classic Nakamoto in three ways: two step transaction confirmation, counting orphaned blocks when estimating network power, and targeting the orphan rate rather than block time. The target rate is 2.5 percent. Because of this the epoch is fixed at 4 hours while the number of blocks inside it floats between 300 and 1,800: with good network connectivity blocks come more often, with poor connectivity less often, and the protocol treats that as normal.

The weak spot of the algorithm is exactly its strength. Eaglesong is used by no other network, there is no merged mining here, and so an ASIC built for it is single purpose: when the coin price falls there is nowhere to point the machine. The second consequence of a narrow market has been visible since April 2024: no new models are being released at all, and the hardware fleet has been frozen in place.

Full description of the Eaglesong algorithm →

Issuance here is dual, and that is the main thing to understand about the economics. Base issuance behaves like Bitcoin: a hard cap of 33,600,000,000 CKB spread over roughly 84 years, until 2103. The reward is set per epoch rather than per block: 1,917,808 CKB per epoch at launch, 958,904 CKB after the first halving. Since an epoch holds between 300 and 1,800 blocks, the actual per block reward drifts, and that is normal.

Secondary issuance works on a completely different principle: exactly 1,344,000,000 CKB per year, forever, with no cap and no halvings. It splits three ways in proportion to how the state space is used: the share equal to occupied state goes to miners, the share equal to NervosDAO deposits goes to depositors, and the remainder is counted as the treasury share and is burned today. The point is that once base issuance runs out, miners still keep an income that does not depend entirely on fees.

From this follows NervosDAO, the most underrated element of the network. A holder who puts CKB into NervosDAO receives the exact portion of secondary issuance that offsets dilution. Put simply, a NervosDAO deposit turns CKB into a hard capped asset, like Bitcoin. Whoever just holds coins, neither in state nor on deposit, slowly pays for their place in the system in favour of miners. That is the state storage fee, collected through inflation rather than through a tax.

A halving happens every 8,760 epochs, and since an epoch is fixed at 4 hours that comes to exactly 4 calendar years, without the drift that Bitcoin shows. The first halving landed on November 19, 2023 at epoch 8,760. The second is expected in November 2027 at epoch 17,520, and the epoch reward will fall to 479,452 CKB. Important: a halving cuts only the base part, secondary issuance stays as it is, so the drop in miner income at a halving is gentler here than in networks with a single reward source.

One more detail that almost every directory gets wrong. The figure of 33.6 billion is not the total supply, it is only the part that will go to miners. Separately, a further 33.6 billion was issued in genesis, and a quarter of that, 8.4 billion, was burned immediately. On top of that comes endless secondary issuance. Total supply today is therefore already around 50 billion CKB and keeps growing.

The asic.es catalogue holds eight Eaglesong models from two vendors, and all eight are covered below: the market is so narrow that there is nothing to filter out.

Browse miners in the catalogue →

The ranking follows catalogue efficiency strictly, from the best joule per terahash to the worst, and the electricity threshold is calculated honestly, without counting the price of the machine itself.

The headline first, with no illusions: Eaglesong machines are not on the HashCore Toolkit support list, and we build no AsicBoost images for this algorithm. Our tooling targets SHA-256 and Scrypt, where the hardware fleet runs to hundreds of thousands of machines and development pays for itself. Eight Eaglesong models in the catalogue is far too narrow a market, and we will not promise firmware that is coming.

A third party firmware market for Eaglesong barely exists either. Neither for the Antminer K7 nor for the Goldshell CK line is there a mature publicly available product comparable to what SHA-256 has. Isolated videos circulate claiming the Antminer K7 can be pushed to 90 TH/s and beyond, but there are no confirmed sources, no public builds and no coherent support behind those claims. Treat them as unverified statements rather than as a ready solution.

The practical conclusion for an Eaglesong ASIC owner: run the official vendor firmware and look for optimisation elsewhere. Three levers are genuinely available: power quality and line stability, intake air temperature, and the choice of a pool with a lower fee. None of them will deliver the sort of jump that a careful undervolt gives on SHA-256, because the platform itself was never built for fine grained frequency and voltage tuning.

A word on risk. Any unverified firmware on a machine drawing three kilowatts is not an experiment, it is a fire load: a mistake in the voltage table kills a hashboard in minutes and voids the vendor warranty. On a narrow market where no new machines have shipped since April 2024, a dead Antminer K7 will be very hard to replace.

The startup routine is standard and takes an evening. You need the machine itself, a dedicated electrical circuit with current headroom, a wallet with a ckb1 address, an account or simply an address on a pool, and a wired network. In the web interface of the machine you enter the pool address, put your CKB address as the worker name, and the password is usually arbitrary. Within a few minutes the first shares show up on the pool dashboard.

Electrics matter more than people expect. The Antminer K7 and full size Goldshell CK6 draw 2.4 to 3.3 kW continuously, which is a current of roughly 11 to 15 amps at 230 volts. Each machine needs its own breaker, an industrial socket and cable of the right cross section. From three machines onward it is worth bringing in a three phase supply and spreading the load across phases, otherwise the imbalance will trip the main breaker.

On noise, honestly. Full size Eaglesong machines are industrial fans at a level where holding a conversation in the next room is uncomfortable. A living space is ruled out entirely. Only the Goldshell CK-BOX and CK-BOX II at 215 and 400 W suit a home, and even then only if you can live with a steady hum. The CK Lite at 1.2 kW is already borderline, more of a garage or a utility room case.

Plan for heat in advance. Three kilowatts going into a machine come back into the room as heat in full, like three powerful heaters with no thermostat. In a Spanish summer that means forced extraction of hot air outdoors is mandatory, while in winter the heat counts as part of the useful output if you were heating that room anyway.

And the main point before buying. At the current CKB price and European tariffs every machine in the catalogue, the Antminer K7 included, runs at a loss on the electricity bill alone. Entering this coin in 2026 makes sense only if you have your own generation, surplus power to dump or a heat reuse task, and not if you are counting on profit.

Miner income here has four parts: base issuance, a share of secondary issuance, 40 percent of the fees for proposed transactions and 60 percent for committed ones. The fee portion on this network is still negligible, so in practice the two issuance components decide everything. The actual block reward at the end of July 2026 held in a range of roughly 540 to 585 CKB, and the spread is explained precisely by the floating number of blocks per epoch.

Three multipliers set the result: the coin price, the network hashrate and your electricity tariff. The price is at an all time low right now, and that makes the economics negative for all eight catalogue models on any European household tariff. Network hashrate has fallen several times over in the last year and a half, and that is the one factor playing in favour of the remaining miners: each machine now takes a bigger share of the reward.

Income should be counted through the electricity threshold, not through a promised payback period. The threshold is the tariff at which the machine breaks even on the power bill, ignoring its purchase price. For the Antminer K7 it stands at about 0.055 dollars per kWh at the end of July 2026, for the whole Goldshell line around 0.014 to 0.016, and for the Antminer K5 around 0.002. The gap between the first and second rows is roughly threefold, and it explains why a conversation about mining CKB today is essentially a conversation about one model.

What would change the picture. A rising price lifts every threshold proportionally: a tripling of the coin would push the Antminer K7 threshold to about 0.16 dollars, which is already comparable to real European tariffs. A further drop in network hashrate also works in favour of those who stay. The second halving in November 2027, on the other hand, will cut the base part of the reward in half, and that belongs in any calculation running longer than a year.

Open the profitability calculator →

A key trait that general reviews skip: on this network it is impossible to hold an arbitrarily small amount on a separate address. Every cell must have a capacity no smaller than the space it occupies together with its scripts, so a minimal working output requires several dozen CKBytes. Pools account for this and set minimum payouts of 100 to 500 CKB. If you send too small an amount to a new address, the transaction simply will not go through.

A second practical point: if you hold what you mined for a long time, it is worth getting to grips with NervosDAO. A deposit offsets that very dilution from secondary issuance, and without it mined coins slowly lose their share of total supply. NervosDAO support is not present in every wallet: of those listed, Neuron and CKBull provide it. Withdrawing from a deposit passes through a waiting phase tied to epochs, so you cannot pull the funds out instantly.

Eaglesong hardware is bought on the used market: factory new machines are practically gone, and the last new model for this algorithm shipped in April 2024. When buying a second hand Antminer K7, ask for video of it running under load for at least an hour, a screenshot of the web interface showing all three hashboards with their temperatures, and photos of the heatsinks. The typical faults on these machines are age related: worn fans, dried out thermal paste and degraded chips on one of the boards.

The CKB coin itself can be bought on the major venues: Binance, KuCoin, Gate, MEXC, CoinEx, Bitrue and a number of regional exchanges. Keep in mind that daily turnover across all venues now runs to single digit millions of dollars, meaning liquidity is thin: a few tens of thousands of dollars already move the order book. Sell what you mine in parts and with limit orders rather than market ones.

Check the exchange lists from older references before you use them. Some venues that aggregators still list have shut down or no longer trade this pair, and in May 2026 one large exchange removed CKB from cross margin trading, leaving spot only. That is a normal sign of a coin with falling turnover, but an unpleasant surprise if you were counting on leverage.

On security, the usual rules with no discounts. Withdraw what you mine from the exchange to your own wallet, and use a hardware wallet for large amounts. Check the first and last characters of a ckb1 address after pasting from the clipboard: clipboard hijacking by malware remains the most common way miners lose funds.

The main application direction since 2024 is RGB++. The protocol binds Bitcoin UTXOs to CKB cells isomorphically: the state of an asset lives in a cell, while the right to dispose of it is proven by spending the corresponding Bitcoin output. The key difference from familiar designs is that there is no bridge, no wrapped tokens and no separate validator set, and therefore none of the usual point of failure such schemes carry.

On top of this sits the Fiber channel network, a payment layer in the logic of Lightning channels but with support for several tokens at once. The testnet started in September 2024, and through 2025 the branch reached versions 0.6.x. Alongside it lives UTXO Stack, a modular kit for assembling layer two solutions on top of Bitcoin that uses CKB as a settlement layer.

A separate branch, and one very characteristic of this network, is Spore and DOB digital objects. Because a CKByte grants the right to occupy a byte of state, an object can be stored entirely on layer one, with no link to an external server and no file sitting somewhere else. Such an object exists for exactly as long as its space is paid for, and it cannot vanish because someone shut down their hosting.

For a miner all of this matters for one reason: real demand on layer one directly affects the miner share of secondary issuance. The more state is actually occupied by cells, the larger the portion of the 1.344 billion CKB per year that goes to miners instead of being burned as the treasury share. State occupancy is still low, and that is a tangible piece of income left on the table.

CKB security rests on the same work economics as any other PoW network, but the scale here is modest. At a hashrate of roughly 70 to 85 PH/s the entire network fleet is the equivalent of a bit over a thousand Antminer K7 class machines. That means the theoretical cost of a 51 percent attack is incomparably lower than on large networks, and a low coin price only pushes that cost down further. The honest conclusion: the network is protected well enough for its size, but no more than that.

Concentration is a separate risk. Per tracker data at the end of July 2026 one pool holds around 40 percent of hashrate and a second about 11 percent. Formally the dangerous threshold is far off, but in practice this means an agreement between two operators could seriously affect the network. For a miner the rule that follows is simple: all else equal, do not pick the biggest pool.

The NC-MAX consensus adds one non obvious defence. Because the protocol counts orphaned blocks when estimating network power and targets that rate specifically, a miner cannot profit from artificially raising the number of orphans. This design closes off part of the selfish mining scenarios that work under classic Nakamoto consensus.

No major protocol level incidents with loss of funds have been publicly recorded across the whole life of the network since 2019. Both hard forks, Mirana in May 2022 and Meepo in July 2025, went to plan with activation epochs announced in advance. We do not quote a public node count here: it could not be obtained from a trusted source on the build date.

The Meepo edition belongs to 2024. The most frequent error. The edition does carry 2024 in its name, but on mainnet it activated on July 1, 2025 at epoch 12,293, while in 2024 it only reached the testnet, on October 25, epoch 9,690. The same goes for the previous edition: Mirana is signed with the year 2021, yet mainnet received it on May 10, 2022 at epoch 5414.

Maximum supply is 33.6 billion CKB. Wrong as a total supply figure. The 33.6 billion is only base issuance, which will go to miners over roughly 84 years. Separately, another 33.6 billion was issued in genesis, of which 8.4 billion was burned immediately, plus endless secondary issuance runs at 1.344 billion a year. Actual total supply is already around 50 billion and rising.

The block reward is 533 CKB. That is the base part, calculated at the maximum of 1,800 blocks per epoch and without secondary issuance. The actual reward at the end of July 2026 held between 540 and 585 CKB across trackers and moves along with the number of blocks per epoch.

Network hashrate is in the hundreds of PH/s. Such figures belong to 2024, when the average sat around 400 PH/s. At the end of July 2026 trackers give 66 to 90 PH/s, a fall of several times over. References quoting hundreds of petahashes describe a state of the network two steps back.

Block time is fixed. It is not. The NC-MAX consensus targets the orphan rate near 2.5 percent rather than block time. The documentation sets a block time range of 8 to 48 seconds, only the epoch length is fixed, at exactly 4 hours, and the number of blocks per epoch floats between 300 and 1,800.

The halving is tied to a block count. Here it is tied to epochs: exactly 8,760 epochs of 4 hours, which is precisely 4 calendar years. That is why halving dates are known in advance and do not drift, unlike networks where the halving is counted by block height.

The halving cuts miner income in half. Only the base part. Secondary issuance of 1.344 billion CKB per year is not touched by a halving at all, so the total drop in the block reward turns out to be well under a half.

The RGB++ protocol launched in April 2024. The sources disagree on dates. Cipher Wang announced the protocol on February 13, 2024, developer tooling and wallets shipped through the whole year, and part of the references place the launch in July 2024. The correct wording in 2026 is simply 2024, with no exact month.

The network launched on November 15, 2019. Official documentation and the project timeline give November 16, 2019. The one day discrepancy in some aggregators comes from the time zone in which they recorded the event.

There is overclocking firmware for Eaglesong. No mature public product exists for this algorithm, neither for the Antminer K7 nor for the Goldshell CK line. Isolated videos claim overclocks to 90 TH/s and above, but no verifiable sources or open builds stand behind them. They cannot be treated as a ready solution.

New Eaglesong ASICs come out regularly. The last new machine for this algorithm shipped in April 2024. From then to the end of July 2026 no further models have reached the market, and Bitmain has not refreshed its line since January 2023.

The Antminer K7 does 66 TH/s. The base specification of the model is 63.5 TH/s at 3,080 W. Sellers also list 58 and 66 TH/s versions, which are different batches and settings rather than one machine. When buying, verify the real hashrate in the interface rather than in the listing text.